Mouse models of genetic prion disease

Over the past few years, as Sonia and I have explained to countless scientists the problem we’re trying to solve, one question we always get is: “is there a mouse model?” The answer is so decidedly yes that it is sometimes difficult to even convey just how rich in mouse models the prion field is. Unlike any other neurodegenerative disease that we know of, prion disease is not just a human disease. Rather, it has several natural mammalian hosts — sheep, goats, mink, deer, elk, cattle. Prions from humans and these other hosts have been transmitted to, and can be serially propagated in, wild-type mice (and hamsters, among other animals). Inoculation of mice with prions produces highly reliable incubation times followed by a steep neurological decline and death closely mimicking the disease course seen in humans or in those other natural hosts. In addition to models using wild-type animals, there are many, many mouse models genetically engineered for heightened susceptibility to prions. I don’t believe that anyone has ever tried to exhaustively catalog all of them, but I would wager that there exist well over 100 different genetically engineered mouse models of prion disease — transgenics expressing all different species of PrP, at all different expression levels, from various transgene constructs or knock-in alleles. These in turn exist on various mouse strain backgrounds, crossed to various lines with endogenous PrP knocked out, and so on and so on.

Yet for what an embarrassment of riches this all is, modeling genetic prion disease in mice has remained a challenge. I was reminded of this last week, when a long-awaited new paper from Joel Watts announced a series of new mouse models of genetic prion disease [Watts 2016]. This prompted me to go back and review the literature to date on modeling genetic prion disease in animals. I reasoned that, thinking from a therapeutics standpoint, it would be really valuable to have mice that reflect true genetic prion disease as closely as possible. For one, because all effective antiprion small molecules discovered to date have proven strain-specific [Kawasaki 2007, Berry 2013, Lu & Giles 2013, Giles 2015, Giles 2016], you’d like to have the exact same prion strain replicating in mice as is the case in humans with that mutation. Another point is that, although we don’t yet know of example, one can imagine there might exist therapeutics that are effective against the spontaneous generation of prions, but not against the replication of prions once they’ve been inoculated. For testing such a therapeutic, mice inoculated with prions just won’t do.

I therefore collected some data from the literature on two possible approaches: (1) transmission of prions from patients with genetic prion disease to mice or other animals, and (2) expression of PrP with pathogenic mutations in mice. Below are two tables summarizing what I found.

First, the transmission of genetic prion disease to animals. This list is not exhaustive, and I welcome suggestions (comments section below) on things that should be added here.

Table 1. Reported findings on the transmissibility of human genetic prion disease caused by various PRNP mutations.

Of note, prions associated with all of the different mutations have proven at least somewhat transmissible, some of the time, to some hosts. But for no mutation is transmissibility guaranteeed. Even for E200K, which I have heard some people opine is indistinguishable from MM1 sCJD, a number of cases have failed to transmit either to primates or to human/mouse chimeric mice. I was surprised, looking back at the literature, how little has been done (or maybe I just missed it?) in terms of transmitting these prions to fully humanized mice (mice expressing human PrP). There are several transmissions to mice expressing MHu2M, a human/mouse chimeric PrP molecule [Telling 1995, Telling 1996b] but fewer reports with mice expressing wild-type human PrP [Telling 1994, Collinge 1995].

The next table reviews reports of mice expressing PrP with mutations that cause genetic prion disease, including the new mice just reported by Watts et al. I tried to make this list exhaustive, so definitely let me know if I have missed anything:

Table 2. Attempts to model spontaneous prion disease caused by genetic mutations in mice. blank = not reported (experiment was either not done or never published); limited = <100% attack rate and/or transmissible only to mice of same genotype or overexpressers; * = susceptible to VV1 sCJD prions; A224V prions not tested

Everything in the above table should be taken with a grain of salt. Characterizing transgenic mice is hard, and there are lots of opportunities to get it wrong. Of course, there is the usual array of questionable research practices that one must always be on the lookout for in science (changing the N or time course after the experiment has begun, post-hoc subgroup analyses, and so on). Then there is also the fact that we’re all human, and even a blinded observer may become biased, thinking that one sick mouse in a cage increases the prior for another mouse in the same cage to be deemed sick. And transgenes can integrate in ways that disrupt other genes, and breeding mice to transmit transgenes can introduce other genetic background confounders in linkage with the transgene. Having multiple transgenic lines expressing the same construct helps, but even then, caution is warranted.

With all that being said, I find the contents of Table 2 pretty interesting. Most of the mutations that people have tried to model in mice are highly penetrant in humans [Minikel 2016], reliably causing spontaneous and uniformly fatal disease. Yet in many cases, these same mutations have failed to cause spontaneous disease at all in mice, for instance in [Asante 2009]. (I applaud Asante et al for publishing their results, as I highly suspect — actually I know — that there exist other transgenic mouse lines that have been created, found to never get sick, and then never published). In other cases, the mutations expressed in mice do cause spontaneous neurological disease, but that disease is not fatal [Jackson 2009] or is only fatal if the protein is expressed above a certain level [Chiesa 1998, Chiesa 2000]. In still other mouse lines, a fatal neurological disease is produced but is found not to be transmissible [Yang 2009], or transmits only to mice of the same genotype or with limited attack rates [Hsiao 1994]. This may make some sense in the case of mutations such as P102L or A117V, which in humans cause a subtype of prion disease that is barely transmissible, but it is more baffling for, say, E200K, which in humans causes formation of a prion strain that is usually readily transmissible (see above). Paradoxically, the mice producing the most readily transmissible prions are those lacking the GPI anchor [Chesebro 2010, Stohr 2011], even though prions from humans with the C-terminal nonsense mutations Y145X and Y163X have not proven transmissible [Tateishi & Kitamoto 1995, Mead 2013].

Perhaps we’re lucky that spontaneous prion disease can even be modeled at all in mice, considering that mice only live 2 or 3 years and most of these mutations take 50 years to cause disease in humans. Still, as noted above, there are therapeutic reasons to want to be able to model the human disease process as closely as possible.

Watts et al reasoned that because expression of bank vole 109I PrP causes spontaneous disease in mice [Watts 2012], this PrP sequence must be intrinsically prone to prion formation, and might therefore be a better starting point for modeling the effects of pathogenic mutations. And as Table 2 shows, the mouse models in the new paper do have some properties that previous models have lacked [Watts 2016]. The prion disease in these mice is spontaneous, fatal, and transmissible, a combination that has been lacking in most other models to date. The disease is also fairly rapid. In Table 2 above, I did not go into detail on the age at which the various mouse lines develop spontaneous disease, in part because there’s too much information (some rows of the above table summarize data from multiple transgenic lines with different times to illness depending on PrP expression level), but many of the previously characterized mouse lines take a year or two to develop illness. The new bank vole D178N mice get sick in as little as 179 days (the Tg15965 line) and E200K in as little as 119 days (the Tg7271 line). And these mice have exceptionally short incubation times if inoculated with brain homogenate from terminally ill mice of the same genotype — on the order of 40-60 days.

The new mice are still not a perfect model of these mutations, however. Although each mutation does give rise to a distinct neuropathological profile, suggesting the initiation of distinct prion strains, the neuropathology is not exactly a deadringer for the corresponding mutations in humans. For instance, the D178N mice didn’t have particularly dire thalamic pathology, as humans with D178N cis-129M tend to do. And in at least one key respect, all of the strains bear a closer resemblance to the prions generated in spontaneously sick mice expressing wild-type bank vole 109I PrP [Watts 2012] than they do to the corresponding human prion strains: these mice all develop an 8 kDa protease-resistant PrP fragment, in contrast to the 19 and 21 kDa fragments typically seen in human D178N cis-129M and E200K prion disease respectively. A final thought is that we believe that there probably exists a small but existent transmission barrier between human prions and bank vole PrP, because when human prions are inoculated into 109M bank voles, the incubation time shortens on second passage [Watts 2014]. That’s one reason why the prions in these mice may not represent the exact same prion strain as seen in humans with the corresponding mutations.

So where does that leave us? If I had a therapeutic candidate with a mechanism of action that I thought might be specific to the generation, rather than propagation, of prions, I’d want to test it in these mice. If I had a therapeutic candidate with a mechanism of action that I was worried might be strain-specific, I might test them in these mice for speed, but I would ultimately still consider my gold standard to be fully humanized mice inoculated with prions from humans with genetic prion disease. But because such prions don’t always transmit efficiently (Table 1), before embarking on an experiment one would want to be careful to pick an isolate known to transmit efficiently to humanized mice. Even then, there does not always exist a humanized mouse with the same mutation to transmit to, so an amino acid sequence difference may be inevitable, at least for now. On account of all of these limitations, I figure no one model reported to date is perfect, and it’s therefore lucky that we have multiple models available to help us make a best guess as to what would really happen in humans. These limitations also highlight that it is important to understand the mechanism of action of therapeutic candidates, to help in predicting how they will translate from models to actual humans.